IEEE Geoscience and Remote Sensing Magazine - March 2020 - 30

the scattering characteristics of DSs and improve the SNR;
however, SBAS reduces the resolution of the image, causing it to lose some detail. In addition to the conventional
SBAS method many improvements have been made. In
[63], PSs are incorporated into the SBAS method; while
using DSs, as many PSs as possible are preserved so as to
improve image details. In [64], the data sets are processed
on single-look and multilook interferograms, respectively, and the results are then fused. This hybrid approach
combines the advantages of the two data sets, which not
only achieves large-scale processing but also improves
quality. More SBAS-based improvements and variants
can be found in [65].
TEMPORARILY COHERENT POINT-INTERFEROMETRIC
SYNTHETIC APERTURE RADAR
TCP-InSAR includes three innovative steps: TCP identification, TCP networks, and TCP least squares estimators [52]. First, interferograms are selected with relatively
short spatial and temporal baselines. TCP is detected
based on the fact that the characteristics of strong scatterers are less sensitive to the window size and oversampling factor used in the image coregistration [66]. TCPInSAR utilizes an interferogram pair to identify TCP,
so both persistent and partially coherent points can be
detected [66], [67]. Second, TCPs are connected by constructing the local Delaunay triangulation network, and
the long arcs (longer than 500 m) are removed to reduce
atmospheric delay and model the relative motion. Third,
the least square method is used to estimate the deformation parameters of each arc. The arcs that have phase
ambiguities are removed according to the least squares
residuals-based outlier detector [61]. Arcs without phase
ambiguities are presented; therefore, it is safe to integrate
the relative parameters with respect to a reference point
to obtain the absolute deformation field. TCP-InSAR has
two distinct advantages: one is that it does not require a
large number of SAR images to perform stable point identification and parameter estimation; the other is that it
uses only arcs without phase ambiguity to solve parameters, which is simple and accurate.

(a)

(b)

STANFORD METHOD FOR PERSISTENT SCATTERERS
In terms of initial PSC selection, StaMPS adopts ADI as a
PSI method; however, the threshold of StaMPS is higher
(lower than 0.4) [41]. Then, based on the spatial coherence
of deformation and atmospheric signals, the model of interferometric phase is established. Using this model, the phase
coherence and elevation correction are estimated iteratively
point by point, and the points with low coherence are eliminated. Therefore, StaMPS, using amplitude and phase, can
identify more stable points than that of PSI using only amplitude, especially in nonurban areas. The elevation correction component is then removed, and phase unwrapping is
performed. The original StaMPS used a 2D phase unwrapping algorithm [41]; later, this method was improved using
a 3D phase unwrapping algorithm [56], [68]. Finally, the
deformation signal is obtained using a filtering operation.
StaMPS has evolved into an open source, time-series InSAR
processing software program of the same name.
SqueeSAR
SqueeSAR processes DSs through spatial adaptive filtering.
The basic principle is as follows.
First, set a fixed-size window to slide in the image. Each
point in the window is tested, with the center point using
the two-sample Kolmogorov-Smirnov (KS) test [69]. Points
that satisfy the same distribution condition are considered
statistically homogeneous pixels (SHPs). Then, the covariance matrix of the DSs is estimated, and a phase statistical
model of the DSs is established. The phase triangulation
algorithm is performed to optimize the phase; therefore,
SqueeSAR can improve the phase SNR without losing the
image resolution. Figure 12 shows the multilook window
and SHPs identified using the KS test. For the estimated
DSs, they can be processed directly in the PSI framework
without the use of additional processing methods. The
earliest DS estimation method based on adaptive window
proposed by SqueeSAR inspired a series of DS-based timeseries InSAR processing techniques. SqueeSAR can achieve
better results than those of traditional PSI methods used
in volcanic deformation, slope monitoring, landslide monitoring [70]-[73], and so forth.

(c)

FIGURE 12. (a) An SAR image, (b) a multilook window, and (c) the identified SHPs (green

points) using the KS test.
30

JOINT-SCATTERER
INTERFEROMETRIC
SYNTHETIC APERTURE RADAR
JSInSAR [46] is another widely used DS
processing method. It includes three
key operations: joint-pixel signal modeling, goodness-of-fit test, and spatial
adaptive filtering. The tested object of
JSInSAR is a joint pixel patch rather
than a single pixel as with SqueeSAR.
This treatment is more resistant to
decorrelation and therefore results in a
more robust estimate. Figure 13 shows
the concept of a joint pixel vector with

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

MARCH 2020

IEEE Geoscience and Remote Sensing Magazine - March 2020

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2020